@Dire,

There are people that are interested in the 1 billion hands. They have 10m already, and that's not enough. Our next goal is to get to 10 billion hands, and our next goal is to have every single hand being played on every poker site.

I will not put anything on any site for any 20$ per month, as I'm not interested in any kind of moneys from any of my internets forums and website activities.

My motivation is to collaborate with people doing research. Some of them write papers (= create ideas and science) and they give us credit for the help they got from us (no matter how big or small).

Having said that - we might very well put up few million of obfuscated hands plus the software (GPL-ed) to parse such proprietary DB. So people will be able to independently build programs that extract and calculate things based on the sample DB. Once they have it running - they can submit the sample to me, and I will execute this over the full DB. With that I would not be the bottleneck for all the things people want to learn and checkout.

The $20 was referred to being paid by you, not end users who would be able to freely access the data.

Hi Dire, I did that.

Checkout http://pokerftp.com

Nothing is yet available for download, but preliminary feedback is welcome.

This is terrific, when do you expect the General Interest samples to be available?

Alright. I have put the downloads, they are now available, and I fixed the site to work with Mozzila and Google chrome.

How many hands are in the sample? Are they all from one site?

You may want to correct the caption on the sample here:
http://pokerftp.com/index_files/examples.htm
for the flop distributions. You noted in this thread that it was actually a lot more hands than noted there, and some breakdown information.

Not many - about 600k. They are from one site. Remember, the point in this sampledb is to test your program for correctness. Then you submit it, and we'll run it against the full DB.

The sample script has "call synch" as the first line and I don't have synch.bat. But the java program works fine.

Synch is leftover from my local version, ignore it. I'm happy to hear that it worked for you, so spread the word and hopefully we get people building nice examples calculating interesting things (And hopefully most of the terrible bugs in the software are gone

I still have not get to implement Aaron's idea on determining the % of winners. If anyone wants to do that, he would be highly welcome.

I am not a coder, but here is a query I have, based on a discussion in the probability forum:

How does allin equity change based on the number of players who VPIP preflop?

I'm interested in the difference between EV win percentage and observed win percentage for people getting allin preflop with the following hands, plotted vs. the number of players who VPIPed before the allin player acted.

low cards: 22, 32, 33, 42, 43, 44, 52, 53, 54, 55
high cards: AA, AK, AQ, AJ, KK, KQ, KJ, QQ, QJ, JJ

Hi,

I'm PeppaPig mentioned in the OP. I was involved in some of the discussions that started this initiative - though I have to say that Indiana has done all the work.

I would have to disagree there. I've spent a lot of time analysing a database with more than 10million hands (but a lot less than 1bn) - and a very common outcome is that there is a fascinating suggestion of some relationship or conclusion - but just not enough hands to prove that its more than noise.

Empirical poker analysis requires a lot of hands because :

a) theres huge variance at every level
b) there are typically lots of dimensions that could be relevant
c) there are often subtle issues of observation bias that need to be adjusted for
d) its very useful to be able to split a large database into multiple independent samples - and verify any conclusion against each sample.

Saying theres nothing to be learned from more than 10m hands is like saying theres nothing to be learned from looking at the night sky with more magnification than binoculars - in fact when you look at the sky through binoculars you can see all kinds of interesting objects that you can't see with the naked eye - but they are all fuzzy and unclear. A 1bn hand database would be equivalent to a large telescope able to show amazing detail of galaxies etc that were just a grey blur with binoculars - and a 10bn hand db would be the Hubble space telescope! In fact - I'm certain that if we had 10bn hands we would still want more!

The huge advantage of empirical results when discussing poker strategy is that it cuts through most of the bs that often comes with this territory - however it does require a huge amount of hands and technology to analyse them to really deliver.

- PeppaPig

Someone needs to build a windows GUI for the java code.

PeppaPig, I don't really think the binoculars example is apt. The thing is when you go from a sample of 10,000,000 to 10,000,000,000 your interval's confidence level just changes from like 99.99% to 99.9999% for most of any practical study. But I am definitely talking about a focused sample. Eg - same stakes, site, game type, general time frame, etc.. I agree that 10,000,000 random hands over random sites in random games/etc could easily be insufficient for alot of things.

Maybe it's kind of ironic that we're debating this anyhow. For 99.99% (har har) of the people who'd be reading any reports, they'd be equally like to flat accept a 95% level over 50,000 as word of god as to say 99% over a billion hands is not a meaningful result since you'd expect to get that out of chance 1 out of a 100 times so it's just running bad. It just depends on what they want to believe. Which was another reason I lost interest in this project. There's oddly enough almost nothing beyond a very rudimentary understanding of statistics amongst most poker players.

@creedofhubris - just to make sure I understand correctly what you want me to do. Is this another "are sites rigged" test? Otherwise why do you want to compare expected v actual equity?

@spadebidder - what should this GUI do?

@Dire - how do you calculate the amount of winning players over a sample of 10m hands? How about "how many hours people play poker on average per day", how do you do that with 10m hands? What about studying how players play, if tags of lags are more profitable? How about collusion amongst players?

Keep in mind that the importance of such database is not only size, but also completeness.